Research On The Curved Image Correction Algorithm And The Multi-scale Robust Matching Algorithm

Image matching is one of the most important tasks in the field of image processing.Image matching technology is playing an important role in the fields of medical treatment,military field,landform detection and transportation safety.Therefore,it is very important to study the new image matching algorithm and improve the performance of classical matching algorithms,so as to improve the applicability and robustness of the image matching algorithm.In this dissertation,an in-depth study of the image correction algorithm and image matching algorithm are implementedIn the first of this dissertation,history,basic concept and research status of image matching technique are briefly introduced,and then the study of image preprocessing and image matching.Before matching the images,it is necessary to preprocess the images.In order to correct the distortion of the curved images,this dissertation proposes a correction algorithm based on the text feature and a correction algorithm of curved images based on polynomial fitting.For the curved image correction algorithm based on the text feature,this paper makes full use of the characteristics of the text image,and the middle lines of the text are used to correct the image,and then completes the correction of the cylindrical text image.For the cylindrical image correction algorithm based on polynomial fitting,the research object is concentrated on the pixel point,and the coordinate relationship between pixel coordinates is explored.Then the image is corrected based on the relationship between pixel coordinates.From the point of view of OCR,the OCR recognition rate is increased from 25.98% to 88.78% through the correction algorithm,and the recognition rate of the corrected document image has been greatly improved compared with the previous image.From the point of view of correlation coefficient,the edge of the image is rectified from curves to straight lines.The experimental results show that the two methods can correct the distorted curved images effectively.The research and analysis object of these two algorithms is the image itself,without taking the relevant parameters of the acquisition device into consideration,so the algorithms have more applicability and the amount of the data that need to be processed is mall.In the research of image matching algorithm,this paper makes an in-depth study of the most widely used algorithm-SURF matching algorithm.SURF algorithm is a matching algorithm with excellent performance at the present stage.Compared with SIFT algorithm,the complexity,robustness and running time of the algorithm are improved greatly.In order to further improve the performance of image matching algorithm,this dissertation proposes a multi-scale robust matching algorithm.At the stage of feature point detection,this paper introduces the mature corner detection algorithm-Harris algorithm,and then combining the advantages of the Harris algorithm and SURF algorithm.At the stage of image matching,the matching strategy based on nearest neighbor algorithm is chosen to match the points.After completing the preliminary matching,using the error matching algorithm to remove the false matching points to improve the number and quality of feature points,and then the feature points that show grate advantages in scale transformation,brightness transformation,noise transformation and rotation transformation can be obtained.The experimental results show that running time of the proposed algorithm can almost remain unchanged,the number of feature points increases greatly,the number of matching points has been improved by about 100 times,and the proposed algorithm is robust and can be used to match the images with higher matching requirements.